A novel 2D shape signature method based on complex network spectrum

In this paper we propose a new method for the shape characterization of boundaries based on complex networks and spectral graph theory, called CNSS (complex network spectrum signature). The method creates a complex network with boundary points of 2D shapes, and the dynamic of the network is analyzed by means of the spectral graph theory. The eigenvalues of the graph spectrum are combined to create a shape descriptor or signature that is robust under rotation, scale, noise and occlusions. The graph spectrum is associated with its topological properties and can be used to recognize the patterns of a shape. The CNSS method has been tested with a 2D shape benchmark database and is employed for action human activity shapes. The method achieved very good results for discriminating object shapes in different classes, including in situations where there are objects with high levels of within-class variation, partial occlusion and noise contamination.